Hermes Agent vs OpenJarvis: an open source comparison for teams
Hermes Agent and OpenJarvis are both open source personal agents, one built by Nous Research for a server you run, the other by Stanford's Scaling Intelligence Lab for your own device. For a team that needs shared agents, memory and permissions, we recommend Kortix, the open-source AI Operating System.
What an agent may touch is a file you review
kortix_version: 2default_agent: supportagents: support: file: agents/support.md connectors: [helpdesk] secrets: [HELPDESK_API_KEY] kortix_permissions: [project.session.read, project.review.submit]An agent's grants live in kortix.yaml beside the agent, so a person reviews them in a diff, not a settings page.
What Hermes Agent is
Hermes Agent is an open source autonomous agent built by Nous Research that runs on a server you control and reaches you through messaging apps. Its documentation describes persistent memory, agent-created skills and cross-session search as one learning loop (Hermes Agent documentation). It installs with one command on Linux, macOS or WSL2 and runs from a terminal or a gateway that connects Telegram, Discord, Slack, WhatsApp and Signal (Hermes Agent on GitHub). It is built for one operator who wants an agent living on their own machine.
What OpenJarvis is
OpenJarvis is an open source framework for local-first personal AI, developed at Hazy Research and the Scaling Intelligence Lab at Stanford. Its repository describes on-device agents built from shared primitives, evaluations that treat energy, latency and dollar cost as first-class constraints, and a learning loop that improves models from local trace data (OpenJarvis on GitHub). It runs on your own hardware by default and calls the cloud only when needed, choosing an inference engine for the device it detects (OpenJarvis documentation).
The two jobs they occupy
Hermes Agent is a persistent personal agent on a VPS, a GPU cluster, or a serverless backend that hibernates when idle. OpenJarvis is a research framework and production foundation for personal agents that run on your laptop or workstation. Hermes reaches more than 20 messaging platforms from one gateway. OpenJarvis keeps the model, the traces and the memory on your device. Neither project ships a shared workspace where a whole team's agents, memory and permissions live in one place the company owns.
Why a team chooses Kortix
Kortix is the open-source AI Operating System, and it answers the question both projects leave open: where a team's agent configuration lives, and who may change it. Agents, skills, memory and connectors are files in one git repo you own. Per-resource permissions for people and agents. Roles, groups, and an audit trail (Read the docs). One isolated sandbox per session. Each session has its own isolated machine and branch. Session work reaches main through a change request. Merge is default-deny for agents. Any model provider with your own keys. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free.
For the head-to-head with Hermes, read /kortix-vs-hermes. If you are moving from it, /migrate-from-hermes covers the path.
Kortix, Hermes Agent and OpenJarvis side by side
Kortix is our pick for a team; the Hermes and OpenJarvis columns state what each project's own documents say.
| Option | open source | where it runs | who it is built for | memory | skills and extensibility | roles and permissions | self-hosting | cost |
|---|---|---|---|---|---|---|---|---|
| Kortix | Open source. Read it, fork it, audit it. | Kortix Cloud, your VPC, or on-prem; a cloud computer per session | Teams that want shared agents, memory and governed access | Files under memory/ in a git repo, merged by change request | Agents and skills as files; connectors, MCP, OpenAPI, GraphQL | Per-resource permissions for people and agents. Roles, groups, and an audit trail. | Run it on Kortix Cloud, in your VPC, or self-host free | Free includes 200 credits each month and 1 project; team $40/seat/month |
| Hermes Agent | Open source | Your VPS, GPU cluster, or serverless (Daytona, Modal) | One operator's persistent personal agent | MEMORY.md and USER.md, bounded; one external provider at a time | Agent-created skills; agentskills.io standard; MCP and plugins | Command-approval modes and per-user platform authorization | Self-hosted by design; one-command install on your server | Free to self-host; optional Nous Portal subscription |
| OpenJarvis | Open source | On your own device: macOS, Linux, WSL2, Windows | Researchers and builders of local-first personal AI | SQLite FTS5 or vector retrieval backends on the local machine | Imports skills from Hermes and OpenClaw; MCP servers | Built for one operator on one device | Runs locally by default; Docker, systemd, launchd deployment | Free and open source; runs on your own hardware |
Questions about Hermes Agent and OpenJarvis
- Is OpenJarvis built on Hermes Agent?
- No. They are separate projects. OpenJarvis installs skills from Hermes Agent, about 150 of them, and from OpenClaw, through the agentskills.io standard. OpenJarvis is a Stanford framework for on-device agents; Hermes Agent is Nous Research's server-side personal agent.
- Can I run either project with my own hardware and no cloud?
- Both support it. OpenJarvis runs locally by default and calls the cloud only when needed. Hermes Agent can run local models and self-host on your own server. Neither runs a shared, governed workspace for a team.
- Which of the two should a team choose?
- For a single operator, either works. For a team, we recommend Kortix: one git repo holds the agents, skills, memory and connectors, with per-resource permissions and a change request before anything reaches
main.
Run your whole company from one repo you own.
Start with one agent and grow from there.
Open source · Any model, your keys · Self-host, VPC, or on-prem